lDDT_align

A tool to align protein structures while maximizing lDDT

Here, the lDDT score is calculated on single-atom amino acid representation (either $C\alpha$ or sidechain centroids). No validation of stereochemical plausibility is performed. The algorithm uses dynamic programming to find the optimal alignment, with some heuristics to speed up computation on larger structures.

Installation

To download and install the software, run:

git clone https://github.com/clami66/lDDT_align.git
cd lDDT_align
python -m pip install .

Usage

lDDT_align [-h] [--thresholds thr [thr ...]] [--inclusion-radius R0] [--atom-type type] [--scale SCALE]
               [--gap-penalty GAP_PEN]
               ref query

Performs structural alignment of two proteins in order to optimize their lDDT

positional arguments:
  ref                   Reference protein PDB
  query                 Query protein PDB

optional arguments:
  -h, --help            show this help message and exit
  --thresholds thr [thr ...], -t thr [thr ...]
                        List of thresholds for lDDT scoring (default: [0.5, 1, 2, 4])
  --inclusion-radius R0, -r0 R0
                        Inclusion radius (default: 15.0)
  --atom-type type, -a type
                        Atom type to calculate distances (choices: {CA, centroid}, default: CA)
  --scale SCALE, -s SCALE
                        Scale factor for the initial alignment (default: 1)
  --prefilter PREFILTER, -f PREFILTER
                        Minimum scaled lDDT score to calculate lDDT on full-scale protein (default: 0.3)
  --gap-penalty GAP_PEN, -g GAP_PEN
                        Penalty to open or extend a gap in the alignment (default: 0.0)

Example: aligning two PDB files

Aligning a CASP14 model to the native PDB structure:

$ lDDT_align test/6t1z.pdb test/T1024TS472_1-D1

Reference: test/6t1z.pdb
Query: test/T1024TS472_1-D1
Total lDDT score: 0.711

Ref.	Score	Query
G	-	-
K	0.64	K
E	0.72	E
D	0.78	D
...
K	-	-
T	-	-
K	-	-

Example: aligning a query against a database of structures

It is also posible to quickly align a query against a reference DB to discover proteins with similar fold. In this case, only the total lDDT is output for each query-reference pair. First, build the DB from a folder of PDB files into a pickle file:

$ python src/lDDT_align/buildDB.py data/scop-pdb/ data/scop-pdb.pkl

Then launch the alignments against the pickle file:

$ lDDT_align data/scop-pdb.pkl test/T1024TS472_1-D1

Reference Target lDDT
d1ejfa_ T1024TS472_1-D1 0.192
d1ekea_ T1024TS472_1-D1 0.243
d1e32a3 T1024TS472_1-D1 0.243
...